working-with-architects
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it ingests and persists data from external profiles to maintain context for its advice.
- Ingestion points: The agent reads context from
.agents/em-context.mdand individual report profiles in.agents/reports/[name].mdas defined inSKILL.md. - Boundary markers: The skill does not define specific delimiters for separating user-provided data from instructions, relying on natural language processing.
- Capability inventory: The skill involves file read and write capabilities to manage these context files within the
.agents/directory. - Sanitization: The instructions include a manual sanitization guideline, directing the agent to save 'stable facts and patterns, not guesses, transient frustration, or unresolved interpretations.'
Audit Metadata